I'm always writting these wrapper classes where I want to selectively want to expose the interface of some of the methods of certain attributes to co the containing object. This can mean I spend a lot of time implementing wrapper methods. That's no good. I wrote a class decorator to make this easy, and I realized it's a perfect complement to the new dataclasses module, though it can also be used with normal classes. I figured I'd check if you're interested in that. The interface looks like this:

>>> from dataclasses import dataclass, acts_like
>>> @acts_like('weight', ['__add__'])
... @acts_like('still_fresh', ['__bool__'])
... @dataclass
... class Spam:
...     weight: int
...     still_fresh: bool
>>> s = Spam(42, False)
>>> s + 3
45
>>> if not s:
...     print('the spam is bad')
the spam is bad

It's a handy way to build objects with composition, but still get some of the benefits of inheritance in a *selective* and *explicite* way.

Here's the code: https://github.com/ninjaaron/cpython/blob/acts_like/Lib/dataclasses.py#L978

May require some addtional twiddling to make it work with frozen dataclasses, but I don't think it should be a problem.

Since this has to wait until 3.8 and I'm swamped with 3.7, this is going to have to wait for me to look at it.

Is there a reason it needs to be part of dataclasses itself? I'd suggest separating it out, putting it on PyPI, and getting some traction there, first.